Lorente-Ramos et al. (2025) Accurate calibration of hydrological models with evolutionary computation multi-method ensembles

Identification

• Journal: Environmental Modelling & Software
• Year: 2025
• Date: 2025-09-27
• Authors: Eugenio Lorente-Ramos, Francisco Gomariz‐Castillo, Francisco Pellicer-Martínez, L. Cornejo-Bueno, Sancho Salcedo‐Sanz
• DOI: 10.1016/j.envsoft.2025.106698

Research Groups

• Department of Signal Processing and Communications, Universidad de Alcalá, Alcalá de Henares, Spain
• Department of Geography, Universidad de Murcia, Instituto Universitario del Agua y del Medio Ambiente, Universidad de Murcia, Campus de Espinardo, Murcia, Spain
• UCAM Universidad Católica de Murcia (UCAM), Murcia, Spain

Short Summary

This study introduces the Dynamic Probabilistic Coral Reefs Optimization algorithm with Substrate Layer (DPCRO-SL) to enhance hydrological model calibration. Applied to the abcd model in two Spanish river basins, DPCRO-SL consistently outperformed the benchmark SCE-UA algorithm, demonstrating superior accuracy and reliability, particularly in test scenarios reflecting future projections.

Objective

• To introduce and evaluate the Dynamic Probabilistic Coral Reefs Optimization algorithm with Substrate Layer (DPCRO-SL), a multi-method ensemble approach, for enhancing hydrological model calibration.
• To compare DPCRO-SL's performance against the benchmark SCE-UA algorithm using the abcd hydrological model (with and without a snow-melt module) in two Spanish river basins (Tagus and Guadalquivir headwaters), testing adaptability with lumped and semi-distributed structures and various calibration strategies.

Study Configuration

• Spatial Scale: Headwaters river basins of two main Spanish rivers: Tagus Headwaters River Basin (THRB) with 4 sub-basins, and Guadalquivir Headwaters River Basin (GHRB) with 5 sub-basins.
• Temporal Scale:
  • Climatic data (precipitation, temperature, potential evapotranspiration): September 1983 to August 2018 (420 monthly values).
  • Observed flow: September 1985 to September 2018 (396 monthly values).
  • Warm-up period: September 1983 to August 1985 (24 months).
  • Calibration period: September 1985 to August 2010 (300 months).
  • Test period: September 2010 to August 2018 (96 months).

Methodology and Data

• Models used:
  • Hydrological Model: abcd model (lumped and semi-distributed) complemented with a snow module based on the Water and Snow Balance Modelling System (WASMOD).
  • Optimization Algorithms: Dynamic Probabilistic Coral Reefs Optimization algorithm with Substrate Layer (DPCRO-SL) and Shuffle Complex Evolution - University of Arizona method (SCE-UA).
  • Calibration Strategies: Single basin calibration, Cascading calibration, Batch calibration, Weighted batch calibration.
  • Performance Metrics: Nash–Sutcliffe Efficiency (NSE), Mean Squared Error (MSE), Kling–Gupta Efficiency (KGE), Percent Bias (Pbias).
• Data sources:
  • Observed flow: Monthly data from the official Spanish Database of ROEA (Red Oficial de Estaciones de Aforo).
  • Climatic data (precipitation, temperature, potential evapotranspiration): Monthly raster maps from the Ministry of Agriculture, Food and Environment.

Main Results

• DPCRO-SL consistently outperformed SCE-UA across all calibration strategies and model structures in both basins, with more pronounced improvements during the test period.
• For the Tagus Headwaters River Basin (THRB) during the test period, using the abcd+snow model with Weighted batch calibration:
  • DPCRO-SL achieved NSE = 0.765, KGE = 0.875, MSE = 171.9, and Pbias = 2.6.
  • SCE-UA achieved NSE = 0.689, KGE = 0.748, MSE = 227.8, and Pbias = -14.8.
• For the Guadalquivir Headwaters River Basin (GHRB) during the test period, using the abcd+snow model with Batch calibration:
  • DPCRO-SL achieved NSE = 0.606, KGE = 0.670, MSE = 130.56, and Pbias = -17.0.
  • SCE-UA achieved NSE = 0.440, KGE = 0.524, MSE = 185.62, and Pbias = -34.2.
• DPCRO-SL demonstrated robust convergence behavior, effectively avoiding local optima and achieving more reliable parameter optimization.
• Both algorithms showed better performance in the test period (2010–2018) compared to the calibration period (1985–2010) for NSE, KGE, and MSE, indicating that the models were not overfitted and are suitable for climate change projections.
• The simpler abcd model without the snow module provided satisfactory performance for the THRB, as snow processes were less influential at the basin outlet.
• The abcd+snow module significantly improved flow predictions for the GHRB, highlighting its importance in regions with substantial climatic and hydrological heterogeneity.
• Weighted batch calibration emerged as the most effective strategy for harmonizing sub-basin dynamics and achieving basin-scale accuracy, particularly in complex basins.

Contributions

• Introduction and comprehensive evaluation of DPCRO-SL, a novel multi-method ensemble optimization algorithm, for hydrological model calibration.
• Demonstration of DPCRO-SL's superior performance and robustness compared to the widely used SCE-UA algorithm across various calibration strategies and model structures.
• Validation of DPCRO-SL's capability for climate change projections by showing better performance in the test period (future-like scenario) compared to the calibration period.
• Provision of insights into the optimal choice of hydrological model complexity (with/without snow module) and calibration strategy based on basin characteristics (homogeneity vs. heterogeneity, snowmelt influence).
• Application and analysis in two climatically variable Mediterranean river basins (Tagus and Guadalquivir headwaters), providing regional hydrological insights.

Funding

• Spanish Agencia Estatal de Investigación, Spain, through Projects TED2021-131066B-I00 and PID2023-150663NB-C21.

Citation

@article{LorenteRamos2025Accurate,
  author = {Lorente-Ramos, Eugenio and Gomariz‐Castillo, Francisco and Pellicer-Martínez, Francisco and Cornejo-Bueno, L. and Pérez-Aracil, Jorge and Salcedo‐Sanz, Sancho},
  title = {Accurate calibration of hydrological models with evolutionary computation multi-method ensembles},
  journal = {Environmental Modelling & Software},
  year = {2025},
  doi = {10.1016/j.envsoft.2025.106698},
  url = {https://doi.org/10.1016/j.envsoft.2025.106698}
}